hash cache tuning

[Andrew Bennetts]

Robert Collins wrote:
[...]
> So I think there are two approaches; one is to probe for whether a stat
> is re-readable accurately, another is to allow users to specify how
> general they want the working tree to be.

I think you can't reliably determine timestamp resolution by stat(2)ing:
http://www.kerneltraffic.org/kernel-traffic/kt20040605_260.html

I'm not sure if that's changed in current linux kernels, but even if it has,
there are presumably still early 2.6 kernels in the wild with this behaviour.

-Andrew.